Zep

Zep is an API and web context engineering and agent memory platform for AI development teams and enterprises building production AI agents and assistants.

Key capabilities include:

• Temporal knowledge graph memory that tracks entities, relationships, and facts over time, with fact invalidation to prevent stale context from influencing responses
• Unified context assembly from multiple sources (chat history, JSON business data, documents, CRM, and app events), formatted into token-efficient context blocks for LLM prompts
• Graph RAG retrieval that combines semantic, keyword, and graph-based search, with sub-200ms (P95) context retrieval and assembly for real-time agents
• Context Templates and custom ontologies to filter by domain and run reproducible evaluations

Pricing: Credit-based pricing starts at $25/month, with a free tier available; enterprise BYOC (Bring Your Own Cloud) options are offered.

Notable for using a temporal knowledge graph approach (powered by the open-source Graphiti engine, with Neo4j support) instead of relying only on vector-based RAG, and for published benchmark results showing strong performance on agent memory evaluations. It also provides an MCP server for integration with Claude, Cursor, and other MCP-compatible clients, and supports multiple LLM providers including OpenAI, Azure OpenAI, Google Gemini, and Anthropic.